gov.nih.mipav.view.renderer.WildMagic.AAM.CDOptimizeBase

Direct Known Subclasses:: CDOptimizeBFGS, CDOptimizeCG, CDOptimizePS, CDOptimizeSA, CDOptimizeSD

public abstract class CDOptimizeBase extends Object

This is the Java modified version of C++ active appearance model API (AAM_API). It is modified with a subset of required functions for automatic MRI prostate segmentation. AAM-API LICENSE - file: license.txt This software is freely available for non-commercial use such as research and education. Please see the full disclaimer below. All publications describing work using this software should cite the reference given below. Copyright (c) 2000-2003 Mikkel B. Stegmann, mbs@imm.dtu.dk IMM, Informatics invalid input: '&' Mathematical Modelling DTU, Technical University of Denmark Richard Petersens Plads, Building 321 DK-2800 Lyngby, Denmark http://www.imm.dtu.dk/~aam/ REFERENCES Please use the reference below, when writing articles, reports etc. where the AAM-API has been used. A draft version the article is available from the homepage. I will be happy to receive pre- or reprints of such articles. /Mikkel ------------- M. B. Stegmann, B. K. Ersboll, R. Larsen, "FAME -- A Flexible Appearance Modelling Environment", IEEE Transactions on Medical Imaging, IEEE, 2003 (to appear) ------------- 3RD PART SOFTWARE The software is partly based on the following libraries: - The Microsoft(tm) Vision Software Developers Kit, VisSDK - LAPACK DISCLAIMER This software is provided 'as-is', without any express or implied warranty. In no event will the author be held liable for any damages arising from the use of this software. Permission is granted to anyone to use this software for any non-commercial purpose, and to alter it, subject to the following restrictions: 1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. 2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software. 3. This notice may not be removed or altered from any source distribution. -- No guarantees of performance accompany this software, nor is any responsibility assumed on the part of the author or IMM. This software is provided by Mikkel B. Stegmann and IMM ``as is'' and any express or implied warranties, including, but not limited to, the implied warranties of merchantability and fitness for a particular purpose are disclaimed. In no event shall IMM or Mikkel B. Stegmann be liable for any direct, indirect, incidental, special, exemplary, or consequential damages (including, but not limited to, procurement of substitute goods or services; loss of use, data, or profits; or business interruption) however caused and on any theory of liability, whether in contract, strict liability, or tort (including negligence or otherwise) arising in any way out of the use of this software, even if advised of the possibility of such damage. $Revision: 1.4 $ $Date: 2003/04/23 14:49:15 $ Optimization base class. This abstract class is the base class for all classes implementing specific optimization procedures. author: Rune Fisker, 26/1-1999

Author:: Ruida Cheng

Field Summary

Fields

Modifier and Type

Field

Description

static int

CentralDifference

int

ENumGrad

static int

eoptBFGS

static int

eoptConjugateGradient

int

EOptMethod

static int

eoptPatternSearch

static int

eoptSimulatedAnnealing

static int

eoptSteepestDescent

static int

eoptUnknown

int

ETermCode

static int

etermConsecMaxStepMax

static int

etermDeltaFuncVal

static int

etermGradTol

static int

etermLineSearch

static int

etermMaxFuncEval

static int

etermMaxIterations

static int

etermNoStop

static int

etermStepTol

static int

etermUnknown

static int

FitLine

static int

ForwardDifference

double

m_dDeltaFuncVal

value for stop criterion: abs(f - fplus) invalid input: '<' m_dDeltaFuncVal

double

m_dEta

machine precision.

double

m_dGradTol

A positive scalar giving the tolerance at which the scaled gradient in considered close enough to zero to terminate the algorithm

double

m_dMachEps

Machine precision

double

m_dMaxStep

A positive scalar giving the maximum allowable scaled steplength at any iteration. maxstep is used to prevent steps that would cause the optimazation algorithm to overflow or leave the domain of interest, as well as to detect divergence.

double

m_dStepTol

A positive scalar giving the tolerance at which the scaled distance between two successive iterated is considered close enough to zero to terminate the algorithm

double

m_dTypF

positive scalar estimating the magnitude of f(x) near he minimizer x-star.

int

m_eNumGrad

methods for estimating the gradient nummericaly

boolean

m_fAnalyticGrad

flag indicating if the gradient is to be calc. analyticaly or nummericaly

boolean

m_fLogFuncValues

Logical variable which determines whether function parameters and corresponding return values should be stored for later analysis.

int

m_iStopCriteria

holds the stop criteria in use, e.g m_iStopCriteria = etermMaxFuncEval | etermMaxIterations

int

m_nConsecMax

Number of conseccutive past steps whose scaled length was equal to maxstep

int

m_nConsecMaxStepMax

max number of conseccutive past steps whose scaled length was equal to maxstep to terminate

int

m_nFDigits

A positive integer specifying the number of reliable digits returned by the objective function FN. fdigits is used to set the parameter n (eta) that is used in the code to specify the relative noise in f(x); the main use of eta is in calculation finite difference step size. eta is set to macheps if fdigits = -1.

int

m_nFuncEval

counter to the number of function evaluations.

int

m_nGradEval

counter to the number of gradient evaluations.

int

m_nIterations

counter to the number of iterations.

private int

m_nMaxFuncEval

limit for number of function evaluations.

private int

m_nMaxIterations

A positive integer specifying the maximum number of iterations that may be performed before the algorithm is halted.

private CDOptimizeFuncBase

m_pFuncEvalBase

pointer to the function to be minimized.

CDVector

m_vFuncVal

private CDVector

m_vMethodPar

special parameters used by each optimization method e.g. step size used for calc. nummerical gradient

CDVector

m_vNFuncEval

Vector<CDVector>

m_vvFuncParm
Constructor Summary

Constructors

Constructor

Description

CDOptimizeBase()

Constructor
Method Summary

Modifier and Type

Method

Description

double

EvalFunction(CDVector x)

function and gradient evaluation methods

void

EvalGradient(CDVector x, CDVector gc, double dFuncVal)

function and gradient evaluation methods

int

ExactLineSearch(CDVector xc, double fc, CDVector g, CDVector p, CDVector xplus, double[] fplus, boolean[] maxtaken)

Perform exact line search Input: xc: parameter fc: function value at xc g: gradient at xc p: search direction Output: xplus: new parameter fplus: function value for new parameter maxtaken: max taken in line search return termination code

CDOptimizeFuncBase

GetFuncEvalBase()

get point to func eval. base.

int

LineSearch(CDVector xc, double fc, CDVector g, CDVector p, CDVector xplus, double[] fplus, boolean[] maxtaken)

Warpper for line search

int

LineSearch(CDVector xc, double fc, CDVector g, CDVector p, CDVector xplus, double[] fplus, boolean[] maxtaken, boolean fSoft)

int

MaxFuncEval()

get the number of evaluations

int

MaxIterations()

get limit for the number of iterations

CDVector

MethodPar()

get numerical gradient function

abstract int

Minimize(CDVector x, CDOptimizeFuncBase pFuncEvalBase)

the Minimize function using analytic gradient

int

MinimizeNum(CDVector x, CDOptimizeFuncBase pFuncEvalBase, CDVector vMethodPar)

String

Name()

name of optimization methode

void

NumGrad(CDVector x, CDVector gradient, double dFuncVal)

calculate nummerical gradient function usingdecided gradient calculation method input: x: parameter dFuncVal: function value in x output: gradient: gradient

int

OptMethod()

name of optimization method

void

SetFuncEvalBase(CDOptimizeFuncBase pFuncEval)

set point to func eval. base.

void

SetMachineEps()

----------------------------[ MachineEps ]---------------------------- Calculate machine epsilon Algorithm A1.3.1 - p. 303 Dennis and Schnabel, Numerical Methods for Unconstrained Optimization and Nonlinear Equations 1983, Prentice-Hall

void

SetMaxFuncEval(int nMaxFuncEval)

set limit for the number of function evaluations

void

SetMaxIterations(int nMaxIterations)

set limit for the number of iterations

void

SetMethodPar(CDVector vMethodPar)

set numerical gradient function

int

SoftLineSearch(CDVector xc, double fc, CDVector g, CDVector sn, CDVector xplus, double[] fplus, boolean[] maxtaken)

Perform line search Given g'p invalid input: '<' 0 and alpha invalid input: '<' 1/2 (alpha = 1e-4 is used), find plus = xc + lambda p,lambda in [0;1], such that f(xplus) invalid input: '<'= f(xc) + alpha * lambda * g'p, using backtracking line search Algorithm A6.3.1 p. 325 Dennis and Schnabel, Numerical Methods for Unconstrained Optimization and Nonlinear Equations 1983, Prentice-Hall Input: xc: parameter fc: function value at xc g: gradient at xc sn: search direction Output: xplus: new parameter fplus: function value for new parameter maxtaken: max taken in line search return termination code

int

UmStop(CDVector x, CDVector xplus, double f, double fplus, CDVector g, int retcode, boolean maxtaken)

Decide wether to terminate minimization Modified version of Algorithm A7.2.1 p. 347 Dennis and Schnabel, Numerical Methods for Unconstrained Optimization and Nonlinear Equations 1983, Prentice-Hall Input: x: parameter xplus: new parameter fplus: function value for new parameter g: gradient at x retcode: return code from line search maxtaken: max taken in line search Output: return termination code

int

UmStop0(CDVector x0, double functionValue, CDVector gradient)

Methods inherited from class java.lang.Object
clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Details
- ENumGrad
  
  public int ENumGrad
- ForwardDifference
  
  public static int ForwardDifference
- CentralDifference
  
  public static int CentralDifference
- FitLine
  
  public static int FitLine
- EOptMethod
  
  public int EOptMethod
- eoptUnknown
  
  public static int eoptUnknown
- eoptSteepestDescent
  
  public static int eoptSteepestDescent
- eoptConjugateGradient
  
  public static int eoptConjugateGradient
- eoptBFGS
  
  public static int eoptBFGS
- eoptSimulatedAnnealing
  
  public static int eoptSimulatedAnnealing
- eoptPatternSearch
  
  public static int eoptPatternSearch
- ETermCode
  
  public int ETermCode
- etermNoStop
  
  public static int etermNoStop
- etermGradTol
  
  public static int etermGradTol
- etermStepTol
  
  public static int etermStepTol
- etermLineSearch
  
  public static int etermLineSearch
- etermConsecMaxStepMax
  
  public static int etermConsecMaxStepMax
- etermDeltaFuncVal
  
  public static int etermDeltaFuncVal
- etermMaxFuncEval
  
  public static int etermMaxFuncEval
- etermMaxIterations
  
  public static int etermMaxIterations
- etermUnknown
  
  public static int etermUnknown
- m_dMachEps
  
  public double m_dMachEps
  
  Machine precision
- m_dTypF
  
  public double m_dTypF
  
  positive scalar estimating the magnitude of f(x) near he minimizer x-star. It is only used in the gradient stopping condition given below. typf should be approximately |f(x*)|
- m_nFDigits
  
  public int m_nFDigits
  
  A positive integer specifying the number of reliable digits returned by the objective function FN. fdigits is used to set the parameter n (eta) that is used in the code to specify the relative noise in f(x); the main use of eta is in calculation finite difference step size. eta is set to macheps if fdigits = -1. If f(x) is suspected to be noisy but the approximate value of fdigits is unknown, it should be estimated be the routine of Hamming[1973] given in Gill, Murray and Wright[1981]
- m_dMaxStep
  
  public double m_dMaxStep
  
  A positive scalar giving the maximum allowable scaled steplength at any iteration. maxstep is used to prevent steps that would cause the optimazation algorithm to overflow or leave the domain of interest, as well as to detect divergence. It should be chosen small enough to prevent the first two of these occurrences but larger than any anticipated reasonable stepsize. The algorithm will halt if it takes steps of length maxstep on m_nConsecMaxStepMax conseccutive iterations
- m_nConsecMax
  
  public int m_nConsecMax
  
  Number of conseccutive past steps whose scaled length was equal to maxstep
- m_dEta
  
  public double m_dEta
  
  machine precision.
- m_fAnalyticGrad
  
  public boolean m_fAnalyticGrad
  
  flag indicating if the gradient is to be calc. analyticaly or nummericaly
- m_eNumGrad
  
  public int m_eNumGrad
  
  methods for estimating the gradient nummericaly
- m_iStopCriteria
  
  public int m_iStopCriteria
  
  holds the stop criteria in use, e.g m_iStopCriteria = etermMaxFuncEval | etermMaxIterations
- m_dGradTol
  
  public double m_dGradTol
  
  A positive scalar giving the tolerance at which the scaled gradient in considered close enough to zero to terminate the algorithm
- m_dStepTol
  
  public double m_dStepTol
  
  A positive scalar giving the tolerance at which the scaled distance between two successive iterated is considered close enough to zero to terminate the algorithm
- m_nConsecMaxStepMax
  
  public int m_nConsecMaxStepMax
  
  max number of conseccutive past steps whose scaled length was equal to maxstep to terminate
- m_dDeltaFuncVal
  
  public double m_dDeltaFuncVal
  
  value for stop criterion: abs(f - fplus) invalid input: '<' m_dDeltaFuncVal
- m_nIterations
  
  public int m_nIterations
  
  counter to the number of iterations.
- m_nFuncEval
  
  public int m_nFuncEval
  
  counter to the number of function evaluations.
- m_nGradEval
  
  public int m_nGradEval
  
  counter to the number of gradient evaluations.
- m_fLogFuncValues
  
  public boolean m_fLogFuncValues
  
  Logical variable which determines whether function parameters and corresponding return values should be stored for later analysis.
- m_vFuncVal
  
  public CDVector m_vFuncVal
- m_vNFuncEval
  
  public CDVector m_vNFuncEval
- m_vvFuncParm
  
  public Vector<CDVector> m_vvFuncParm
- m_pFuncEvalBase
  
  private CDOptimizeFuncBase m_pFuncEvalBase
  
  pointer to the function to be minimized.
- m_vMethodPar
  
  private CDVector m_vMethodPar
  
  special parameters used by each optimization method e.g. step size used for calc. nummerical gradient
- m_nMaxIterations
  
  private int m_nMaxIterations
  
  A positive integer specifying the maximum number of iterations that may be performed before the algorithm is halted. Appropriate values depend strongly on the dimension and difficulty of the problem, and the cost of evaluating the nonlinear function.
- m_nMaxFuncEval
  
  private int m_nMaxFuncEval
  
  limit for number of function evaluations.
Constructor Details
- CDOptimizeBase
  
  public CDOptimizeBase()
  
  Constructor
Method Details
- Minimize
  
  public abstract int Minimize(CDVector x, CDOptimizeFuncBase pFuncEvalBase)
  
  the Minimize function using analytic gradient
  
  Parameters:
  
  x -
  
  pFuncEvalBase -
  
  Returns:
- Name
  
  public String Name()
  
  name of optimization methode
  
  Returns:
- OptMethod
  
  public int OptMethod()
  
  name of optimization method
- EvalFunction
  
  public double EvalFunction(CDVector x)
  
  function and gradient evaluation methods
  
  Parameters:
  
  x -
  
  Returns:
- EvalGradient
  
  public void EvalGradient(CDVector x, CDVector gc, double dFuncVal)
  
  function and gradient evaluation methods
  
  Parameters:
  
  x -
  
  gc -
  
  dFuncVal -
- MethodPar
  
  public CDVector MethodPar()
  
  get numerical gradient function
  
  Returns:
- SetMethodPar
  
  public void SetMethodPar(CDVector vMethodPar)
  
  set numerical gradient function
  
  Parameters:
  
  vMethodPar -
- MaxIterations
  
  public int MaxIterations()
  
  get limit for the number of iterations
  
  Returns:
- MaxFuncEval
  
  public int MaxFuncEval()
  
  get the number of evaluations
  
  Returns:
- SetFuncEvalBase
  
  public void SetFuncEvalBase(CDOptimizeFuncBase pFuncEval)
  
  set point to func eval. base.
  
  Parameters:
  
  pFuncEval -
- GetFuncEvalBase
  
  public CDOptimizeFuncBase GetFuncEvalBase()
  
  get point to func eval. base.
  
  Returns:
- SetMaxIterations
  
  public void SetMaxIterations(int nMaxIterations)
  
  set limit for the number of iterations
  
  Parameters:
  
  nMaxIterations - max iterations
- SetMaxFuncEval
  
  public void SetMaxFuncEval(int nMaxFuncEval)
  
  set limit for the number of function evaluations
  
  Parameters:
  
  nMaxFuncEval - max number of evaluations.
- SetMachineEps
  
  public void SetMachineEps()
  
  ----------------------------[ MachineEps ]---------------------------- Calculate machine epsilon Algorithm A1.3.1 - p. 303 Dennis and Schnabel, Numerical Methods for Unconstrained Optimization and Nonlinear Equations 1983, Prentice-Hall
- UmStop0
  
  public int UmStop0(CDVector x0, double functionValue, CDVector gradient)
- UmStop
  
  public int UmStop(CDVector x, CDVector xplus, double f, double fplus, CDVector g, int retcode, boolean maxtaken)
  
  Decide wether to terminate minimization Modified version of Algorithm A7.2.1 p. 347 Dennis and Schnabel, Numerical Methods for Unconstrained Optimization and Nonlinear Equations 1983, Prentice-Hall Input: x: parameter xplus: new parameter fplus: function value for new parameter g: gradient at x retcode: return code from line search maxtaken: max taken in line search Output: return termination code
- LineSearch
  
  public int LineSearch(CDVector xc, double fc, CDVector g, CDVector p, CDVector xplus, double[] fplus, boolean[] maxtaken)
  
  Warpper for line search
  
  Parameters:
  
  xc - parameter
  
  fc - function value at xc
  
  g - gradient at xc
  
  p - search direction
  
  xplus - new parameter
  
  fplus - function value for new parameter
  
  maxtaken - max taken in line search
  
  Returns:
  
  termination code
- LineSearch
  
  public int LineSearch(CDVector xc, double fc, CDVector g, CDVector p, CDVector xplus, double[] fplus, boolean[] maxtaken, boolean fSoft)
- ExactLineSearch
  
  public int ExactLineSearch(CDVector xc, double fc, CDVector g, CDVector p, CDVector xplus, double[] fplus, boolean[] maxtaken)
  
  Perform exact line search Input: xc: parameter fc: function value at xc g: gradient at xc p: search direction Output: xplus: new parameter fplus: function value for new parameter maxtaken: max taken in line search return termination code
- SoftLineSearch
  
  public int SoftLineSearch(CDVector xc, double fc, CDVector g, CDVector sn, CDVector xplus, double[] fplus, boolean[] maxtaken)
  
  Perform line search Given g'p invalid input: '<' 0 and alpha invalid input: '<' 1/2 (alpha = 1e-4 is used), find plus = xc + lambda p,lambda in [0;1], such that f(xplus) invalid input: '<'= f(xc) + alpha * lambda * g'p, using backtracking line search Algorithm A6.3.1 p. 325 Dennis and Schnabel, Numerical Methods for Unconstrained Optimization and Nonlinear Equations 1983, Prentice-Hall Input: xc: parameter fc: function value at xc g: gradient at xc sn: search direction Output: xplus: new parameter fplus: function value for new parameter maxtaken: max taken in line search return termination code
- NumGrad
  
  public void NumGrad(CDVector x, CDVector gradient, double dFuncVal)
  
  calculate nummerical gradient function usingdecided gradient calculation method input: x: parameter dFuncVal: function value in x output: gradient: gradient
- MinimizeNum
  
  public int MinimizeNum(CDVector x, CDOptimizeFuncBase pFuncEvalBase, CDVector vMethodPar)

Class CDOptimizeBase

Field Summary

Constructor Summary

Method Summary

Methods inherited from class java.lang.Object

Field Details

ENumGrad

ForwardDifference

CentralDifference

FitLine

EOptMethod

eoptUnknown

eoptSteepestDescent

eoptConjugateGradient

eoptBFGS

eoptSimulatedAnnealing

eoptPatternSearch

ETermCode

etermNoStop

etermGradTol

etermStepTol

etermLineSearch

etermConsecMaxStepMax

etermDeltaFuncVal

etermMaxFuncEval

etermMaxIterations

etermUnknown

m_dMachEps

m_dTypF

m_nFDigits

m_dMaxStep

m_nConsecMax

m_dEta

m_fAnalyticGrad

m_eNumGrad

m_iStopCriteria

m_dGradTol

m_dStepTol

m_nConsecMaxStepMax

m_dDeltaFuncVal

m_nIterations

m_nFuncEval

m_nGradEval

m_fLogFuncValues

m_vFuncVal

m_vNFuncEval

m_vvFuncParm

m_pFuncEvalBase

m_vMethodPar

m_nMaxIterations

m_nMaxFuncEval

Constructor Details

CDOptimizeBase

Method Details

Minimize

Name

OptMethod

EvalFunction

EvalGradient

MethodPar

SetMethodPar

MaxIterations

MaxFuncEval

SetFuncEvalBase

GetFuncEvalBase

SetMaxIterations

SetMaxFuncEval

SetMachineEps

UmStop0

UmStop

LineSearch

LineSearch

ExactLineSearch

SoftLineSearch

NumGrad

MinimizeNum